NeuroCOLT

Neural Networks and Computational Learning Theory

 

About NeuroCOLT

Papers Archive

1994 1995
1996 1997
1998 1999
2000 2001

Books

info@neurocolt.org

NeuroCOLT Technical Report NC-TR-98-009



Learning via Internal Representation
(Extended Abstract)


Eli Dichterman
LSE & RHUL
UK

Keywords: Internal representation; learning framework

Received: 11-MAY-1998


Abstract
We present a learning framework based on reducing a learning task to the problem of finding a good internal representation of the input examples; a good internal representation is a set of features, relative to which a simple generalization rule, such as a linear hyperplane classifier, can be applied to obtain a good approximation of the target. Finding a good internal representation of a learning problem is especially useful when the same representation is good for a set of similar tasks, such as the recognition of several different characters. Although the problem of extracting a set of informative features is not easier, in general, than the learning problem itself, we show that some of the most effective learning mechanisms, such as Boosting and Support Vector Machine, are actually based on efficient methods of extracting good internal representations of the input data. In particular, we analyze and correct a recent approach based on approximately embedding a distribution-depended learning task in a high-dimensional Euclidean space.

Download Compressed Postscript